More Americans are asking AI for money advice – and revealing too much personal information. What you should not share

Would you instrumentality fiscal proposal from a sociopath? If you're among the bulk of Americans who usage AI for wealth management, it's apt you already have.

MIT concern prof Andrew Lo co-authored a 2023 insubstantial that equated ample connection models (LLMs) similar ChatGPT to "a quality sociopath" owed to their inability to explicit empathy (1).

Yet that doesn't look to fuss the 55% of Americans who told a March 2026 TD Bank survey that they "use AI to assistance negociate their finances (2)." That's up from conscionable 10% the twelvemonth before.

Must Read

• Thanks to Jeff Bezos, you tin present go a landlord for arsenic small arsenic $100 — and no, you don't person to woody with tenants oregon hole freezers. Here's how

• Robert Kiyosaki says this 1 plus volition surge 400% successful a twelvemonth and begs investors not to miss this ‘explosion’

• Dave Ramsey warns astir 50% of Americans are making 1 large Social Security mistake — here’s however to hole it ASAP

Of course, utilizing AI for fiscal assistance isn't new. Americans crook to it for everything from fiscal literacy questions to savings plans, banal marketplace tips and status proposal (3). Lo himself has spoken astir however to usage AI arsenic a fiscal advisor. The rising interest astir AI and finances, however, is much astir what accusation radical are providing LLMs to unafraid the proposal they're looking for.

A 2024 Cisco survey confirmed that up to 29% of AI users input relationship numbers and different fiscal accusation — contempt the immense bulk acknowledging that their information could beryllium shared (4).

But the sharing of information isn't the lone concern. Thieves person discovered devious ways to assistance your idiosyncratic info from LLMs — information that they tin past usage to bargain your individuality and your money.

How criminals bargain your information from AI tools

Last year, Stanford University researchers studied six large U.S. LLMs: Nova, Claude, Gemini, Meta AI, Copilot, and ChatGPT (5).

Lead writer Jennifer King said they recovered that immoderate "sensitive information" shared with specified LLMs "may beryllium collected and utilized for training" — a large problem, she added, due to the fact that "almost nary probe has been conducted to analyse the privateness practices for these emerging tools."

NordPass, a part of Nord Security, warned that information breaches astatine LLMs tin alteration "malicious actors" to "view your implicit chat history, including immoderate delicate information that you shared with the AI instrumentality (6)."

Others, meanwhile, enactment that idiosyncratic accusation you upload is utilized to bid AI models, which past "could reproduce verbatim text, exposing idiosyncratic accusation (7)."